The AI ​​tool dazzles as much as it worries

His name spread like wildfire within a few weeks. It was launched at the end of November. Chatbot ChatGPT Detailed answers to questions on a wide range of topics can be formulated. And the capabilities of this tool to California startup OpenAI, which has been “trained” thanks to the vast amounts of data gleaned from the Internet, makes you dizzy. Able to write a poem, respond to a philosophical topic, propagate a scientific idea to a five-year-old, create a recipe with what’s left in the fridge or even write a complex program of code – all in seconds -, This artificial intelligence It fascinates as much as it worries.

“It is the largest model in existence in terms of the parameters and data used. And from a technical point of view, it is undoubtedly the most efficient model,” says Mary Alice Plate, a data engineer specializing in artificial intelligence. And if the media machine around chat It has been deprecated, as it is also the first chatbot accessible to the general public. Usually, advances in artificial intelligence remain in the scientific field. There, everyone could use the interface, and everyone did. There was a snowball effect, it created a real simulation, ”adds the specialist.

No source, no reliability

But like a romantic relationship, once the honeymoon is over, the skies get dark. After several weeks of excitement, some professionals are warning about the reliability of the answers ChatGPT provides. “It’s a script generator that works well, but it doesn’t guarantee that the information provided is correct,” says Amélie Cordier, Scientific Director of Once for All.

First, because the data built into the tool stops at 2021. Then, because this chatbot can’t search directly on the web, explains Virginie Mathivett, director of data science and engineering at TeamWork. The tool doesn’t incorporate data from the last few months, nor does it update. So if you ask him who made a profit soccer world cup in qatar, He will not be able to answer.” In other cases, the answers he will give may be wrong, warns Mary Alice Plate: “They are misleading. I took the test by asking him a question about Pension reform. His interpretation was correct, but the answer was ultimately wrong because it was based on numbers from 2021.”

Especially since the bot explains its answers without mentioning its sources. “It’s an accurate reflection of the internet. And on the internet, there’s everything, trusted sites as well as untrusted sites.” “But when you do a Google search, you can quickly find out whether the site you are looking for is authoritative or not. There, it is impossible to know the source of the information given by the tool,” Mary Alice Blitt continues. As the expert reminds us, the goal of ChatGPT is not “to give the best answer to the question, the most factually accurate, but the most reasonable answer that can be found on the Internet”.

Drafting challenge

To avoid the spread of fake news, the California startup has put in place safeguards on certain topics, according to the specialist: “If you ask questions about climate, you will get answers that are not climate-sceptical. But on other topics, less present, this can lead to the emergence of fake news “.

What worries professionals the most is that the answers are not the same depending on how you manage your question. The tool asked about ways to treat depression. I received a detailed answer, with acceptable explanations. Then I asked him how electric shocks are a good way to treat depression. And I already got an answer that shows me that it’s a good way. Virginie Mathewett warns of the danger if the questions are biased.

For Katya Lainé, co-founder and CEO of Talkr.ai — a French independent publisher and resource of bot technology, platform, and conversational AI — the challenge is teaching the public to use the tool. “It’s like any tool, you have to know how to use it. To drive a car, you have to pass the license before, right there, you have to know how to use it,” she adds. For a poem, recipe or email, it poses some problems, explains the specialist, but you need to be especially careful with scientific or medical questions: “He may have the right answer, but it’s not automatic. It’s absolutely essential to check the information with a reliable source.” “.

necessary adjustments

The first objectives of this council are students and pupils. Just a few weeks after its launch, ChatGPT’s impact on the education world is already being felt. Fearing a wave of fraudEight Australian universities, especially with regard to “homework”, have announced that they will adjust their exams, citing a ban on the use of artificial intelligence by students. Because the tool is able to develop a dissertation on any topic from quantum physics to Scandinavian literature, and produces “unique” texts. In other words, it is impossible for two students to submit the same assignment, which makes it difficult for teachers to detect the use of ChatGPT. “If only one student used it, it would be difficult to tell. But if ten students used it, even if they did not have the same copy, they would be similar in their construction,” says Mary Alice Plate.

And you’ll feel the limits very quickly, Virgin Mathewett believes: “It can help or guide a student with his homework, but it won’t be enough for everything you teach. It’s a tool like Wikipedia Or Google.” In the 2000s, the same fear was expressed with the arrival of Wikipedia, as Amelie Cordier recalls: “Today, all information is at your fingertips. Teaching must adapt to the tools available to students. And for students to learn how to use it and spot the risks.”

For an expert, whether on the teaching side or in other fields, this robot – and artificial intelligence as a whole – will necessarily cause disruptions. “Some professions will be forced to adapt, but that’s not necessarily a bad thing. When Excel arrived, it didn’t replace accountants, they just adapted.”